For all the talk of AI in medicine, most of it still happens behind the scenes — sorting scans, drafting notes, flagging risks. A personalized cancer vaccine called EVX-01 is different: here, the AI didn’t assist the treatment, it designed it. And in melanoma patients, that AI-designed vaccine is showing results strong enough to take seriously.
What the AI actually did
EVX-01 is made by Evaxion using its AI-Immunology platform. For each patient, the system analyzes the unique mutations in their tumor and predicts which of them — called neoantigens — the immune system is most likely to recognize and attack. Those predictions become the personalized targets built into the vaccine. It is precision medicine in the most literal sense: a different vaccine for every patient, designed by an algorithm reading that patient’s own cancer.
The results
In a phase 2 trial, EVX-01 was given alongside the established immunotherapy pembrolizumab (Keytruda) in patients with advanced melanoma. The combination produced a 75% objective response rate — 12 of 16 patients showed a clear clinical response — and 11 of those 12 responders were still responding at the two-year follow-up. For advanced melanoma, durable responses like that are exactly what doctors hope to see.
Proof the AI picked the right targets
Newer data presented at the American Association for Cancer Research annual meeting in April 2026 added a crucial detail: 86% of the vaccine targets the AI selected actually triggered a tumor-specific immune response — a record-high precision rate. That number matters because it validates the hardest part of the approach. The AI isn’t just generating plausible-looking targets; the targets it chose are the ones the body genuinely fights.
Why this is a milestone
Personalized cancer vaccines have been a tantalizing idea for years, but the bottleneck has always been picking the right targets fast enough and accurately enough to matter. That is precisely the problem AI is suited to solve. EVX-01 is one of the first real-world demonstrations that an AI-designed therapy can move from prediction to patient and produce durable clinical benefit.
The honest caveats
This is still a relatively small trial, not an approved treatment, and bigger studies are needed before EVX-01 reaches the clinic broadly. Early results can soften at scale, and a 16-patient response rate is a promising signal, not a guarantee. But the direction is clear, and the validation of the AI’s target selection is the kind of evidence that turns a hopeful idea into a real research program.
The bottom line
EVX-01 shows AI doing something genuinely new in medicine: not summarizing or assisting, but designing a treatment tailored to a single patient’s tumor — and seeing it work. It is an early but real glimpse of how artificial intelligence could reshape cancer care, one personalized vaccine at a time.
